摘要
提出了一种改进的支持向量分类方法,根据支持向量机中支持向量不会出现在两类样本集间隔以外的正确划分区的理论,通过引入类质心距等概念,从而较好地解决了当两类样本集混淆严重的时候如何更加精确地进行剔除混淆点,保证算法泛化性的问题。实验表明,采用这种改进的算法在两类训练样本集混淆较严重时能较好地解决泛化性问题。
A new method of Support Vector Machine（SVM） is presented in this paper. Because support vector will not appear in the area which is out of inteval between two classes ,the algorithm introduces some new concepts such as class - centroid, class - radius class -centroid -distance. With these concepts we can delete those test samples which are not Supprot Vectors（SV） quickly and exactly. We also improve K - Nearest Neighbour（KNN）. With the new concept named class - centripetal force, we solve problem that delete promiscuous test example exactly, and keep the generalization. The experiments show that the advanced algorithm can achieve the excepted target.
出处
《现代电子技术》
2007年第1期150-152,共3页
Modern Electronics Technique
关键词
支持向量机
类向心度
样本集
KNN
support vector machine
class - centroid - distance
sample classes
KNN
